Problem
The supabase-postgres-best-practices skill ships with CLAUDE.md and SKILL.md that are ~90% identical (both 65 lines, same content). Claude Code loads CLAUDE.md into every conversation context automatically. Since SKILL.md already contains the full guidelines and routing table to references/, the duplication wastes context window tokens on every session.
Current state
CLAUDE.md: 69 lines — full description, priority table, usage instructions, referencesSKILL.md: 65 lines — nearly identical contentreferences/: 31 on-demand rule files (well-structured, no issue here)
Suggestion
Slim CLAUDE.md down to a short pointer:
# Supabase Postgres Best Practices
Postgres optimization skill from Supabase. See `SKILL.md` for full guidelines and `references/` for detailed rules (loaded on-demand).
This preserves discoverability while saving ~60 lines of duplicated context per conversation. The SKILL.md and references/ structure is already excellent — the issue is just the redundant CLAUDE.md.
Context
This matters because CLAUDE.md is injected into the context window at conversation start. For users with multiple skills installed, every duplicated line compounds. The Railway skill (railwayapp/railway-skills) handles this well — no CLAUDE.md at all, just a well-structured SKILL.md with on-demand references.
Filed after installing via npx skills add supabase/agent-skills@supabase-postgres-best-practices
Problem
The
supabase-postgres-best-practicesskill ships withCLAUDE.mdandSKILL.mdthat are ~90% identical (both 65 lines, same content). Claude Code loadsCLAUDE.mdinto every conversation context automatically. SinceSKILL.mdalready contains the full guidelines and routing table toreferences/, the duplication wastes context window tokens on every session.Current state
CLAUDE.md: 69 lines — full description, priority table, usage instructions, referencesSKILL.md: 65 lines — nearly identical contentreferences/: 31 on-demand rule files (well-structured, no issue here)Suggestion
Slim
CLAUDE.mddown to a short pointer:This preserves discoverability while saving ~60 lines of duplicated context per conversation. The
SKILL.mdandreferences/structure is already excellent — the issue is just the redundant CLAUDE.md.Context
This matters because CLAUDE.md is injected into the context window at conversation start. For users with multiple skills installed, every duplicated line compounds. The Railway skill (
railwayapp/railway-skills) handles this well — no CLAUDE.md at all, just a well-structured SKILL.md with on-demand references.Filed after installing via
npx skills add supabase/agent-skills@supabase-postgres-best-practices